Longitudinal clinical validation of the Northstar response using methylated ctDNA as a dynamic treatment response biomarker in advanced stage gastrointestinal (GI) cancers.

Abstract

237 Background: Radiographs remain the cornerstone of tumor response evaluation for managing advanced GI cancers. However, imaging captures only macroscopic bulk tumor burden changes. Liquid biopsies may detect response earlier and more precisely to inform treatment decisions. In particular, methylated circulating tumor DNA (mctDNA) holds promise as a dynamic measurement capturing pan-GI cancer treatment response. Methods: This prospective cohort study enrolled patients with varied advanced GI malignancies receiving systemic therapy from 5/2023-5/2025. Plasma was collected at baseline and before each treatment cycle. These samples were then sent to a CLIA-certified laboratory (BillionToOne, Inc.) for mctDNA quantification using Tumor Methylation Score (TMS). The correlation between TMS and radiographic measures were evaluated using linear plotting of lesions Sum of Diameters (SOD) along with the Log10 change in TMS. The primary objective was to evaluate the correlation between TMS and radiographic measures. Secondary objectives included lead-time between molecular and radiologic response, and TMS performance in informing PFS and OS. Cox proportional hazard models, KM curves, and Kruskal-Wallis testing were used to analyze the endpoints. Results: A total of 106 patients were enrolled, with 81 reaching the first standard of care treatment response timepoint. This included HCC at 27.4% (n=29), CRC 26.4% (n=28), biliary 17% (n=18), pancreatic 12.3% (n=13), esophagogastric (EG) 12.3% (n=13), anal cancer 0.9% (n=1), and other GI cancers 3.8% (n=4). The highest baseline median log10 of TMS was in the “other” GI cancer group (4.6, Interquartile range [IQR] [3.3–5.2]), followed by CRC (4.4, [IQR, 3.6–5.0]). At baseline, patients with both primary and metastatic lesions (n = 44) had a higher median log10 TMS of 3.8 [IQR, 2.8–4.4]. Patients with only metastatic lesions (n = 32) had a median of 2.8 [IQR, 2.2–3.9], while patients with only primary lesions (n = 4) had a median of 2.4 [IQR, 1.3–3.6]. When adjusted for age, tumor types, treatment types, and Charleston Comorbidity Index, a log10 change in TMS as a continuous variable was significantly associated with OS (HR 4.03; p<0.01) and PFS (HR 2.07; p<0.01). Among the initial evaluable cases, the SOD of target lesions were associated with TMS values. Molecular RECIST calls based on TMS dynamics demonstrated changes ahead of RECIST assessments in the initial set of patients. Additional analyses will be available at the meeting. Conclusions: Pan-cancer, pan-therapy plasma mctDNA quantification demonstrates significant potential to predict OS and progression by monitoring changes over time. TMS correlated with disease burden and RECIST qualifying changes. These findings support continued exploration as a tool for preemptive therapeutic decision-making.